Journal of Sound and <ibration (2000) 235(3), 477}494

doi:10.1006/jsvi.2000.2945, available online at http://www.idealibrary.com on

DESIGN OPTIMIZATION OF MARINE ENGINE–MOUNT

SYSTEM
J. S. TAO, G. R. LIU AND K. Y. LAM
Institute of High Performance Computing, 89-C Science Park Drive C02-11/12, ¹he Rutherford,
Singapore 118261, Singapore. E-mail: hpctaojs@nus.edu.sg

(Received 9 December 1999, and in ,nal form 7 March 2000)

Design optimization of marine engine}mount system for vibration control is presented in

this paper. The engine is modelled as a rigid body with supports connected to a rigid #oor.
The mounts are modelled as three-dimensional isolators with hysteresis damping. The
objective is to select the sti!ness coe$cient and orientations of individual mount in order to
minimize the vertical force transmitted from the engine to the #oor to control the structure-
borne noise. Constraints are imposed to keep the isolator static and dynamic de#ection
within the desired limits and a minimum gap between system natural frequency and engine
excitation frequency for avoiding possible system resonance. The sequential quadratic
programming (SQP) technique has been applied as the optimization algorithm. The typical
force and moment of a 4-stroke engine with one cylinder is analyzed and input in the
optimization system. The results of optimization are compared with that of a conventional
engineering design with the isolators working under their maximum allowable de#ection.
This comparison shows that, as compared with force transmission of the conventional
isolation system, the value of optimized system is only half for multi-frequencies excitation
and one-fourth for single frequency excitation. The mechanism of vibration isolation
involved in the optimization is investigated with the frequency response of the system.
Sensitivity study of the system with the variation of the design parameters is also carried out
and proves that the minimum obtained is the global value in the range of design parameters.
 2000 Academic Press.

1. INTRODUCTION

Vibration isolation of marine engine on compliant mounts is a very common and important
engineering design problem. The challenge for the design engineer is how to select the
vibration isolators and how to properly install them in order to minimize the structure-
borne noise and vibration level in the cabins or noise breakout into the water which is
especially critical for warships. Unlike in the normal application in the building, a mounting
system for marine engine should be able to react the strong dynamic force caused by wave
slap, cornering loads, and docking impact. Thus, the static de#ection of isolators shall
be designed within a limit for keeping the balance of engine under the strong impact. This
will increase the vibration transmitted into the hull structure and challenge the design
engineer to create even more e!ective mounting system. One useful method for helping the
design engineer to develop more e!ective mounting systems is through optimization
techniques.
Many researchers have involved in the study of this area. Sevin et al. [1] summarized
some achievements by 1971. However, due to the limitation of computational power,
most of work has been done at that time concentrated on the theoretical analyses and

0022-460X/00/330477#18 $35.00/0  2000 Academic Press

478 J. S. TAO E¹ A¸.

single-degree-of-freedom (d.o.f.) systems. Recently, due to the rapid development of the

computational technology, this area attracts more attention. Spiekermann et al. [2] used
the penalty function optimization method to obtain optimal design of engine mounting in
automobile application by moving six system natural frequencies away from engine
excitation frequency. This function penalized the natural frequencies in an undesirable
frequency range and also large design changes. However, Swanson et al. [3] showed that the
transmitted forces should be directly minimized rather than the natural frequencies in order
to determine a `trulya optimal design of the mount. He used the recursive quadratic
programming technique for the optimal isolation system design of aircraft engine. The
objective function in his study is the total force transmitted to the structure, and the design
variables are the sti!ness of the mounts. The constraint is the dynamic de#ection of c.g of
engine. Ashra#iuon [4] continued Swanson's study but with more design variables and
a #exible base. Snyman et al. [5] carried out another study to minimize the total force
transmitted to the base by optimizing the balancing mass and associated phase angles of
a V-engine. No constraint is imposed in his study.
In summary, the previous studies on the optimization of engine mounting systems mainly
focused on two aspects; One is to minimize the transmitted force by moving the system
natural frequencies away from an undesired frequency; the other is to directly optimize the
force transmission by adjusting the design parameters. There is no report on the theoretical
analyses of the objective function and sensitivity study of their optimization results. It is
possible that the minimum value obtained in their studies is at the anti-resonance point with
system frequencies quite near the excitation frequencies. Harris et al. [6] present the
frequency response of a resiliently supported rigid body under the excitation of force and
moment. It is clearly indicated that there is an anti-resonance frequency between two
natural frequencies of the system. The small shift of the system variables may cause
a resonance of the system and thus totally destroy the initial purpose. On the other hand,
the objective function in their study is the total force transmitted to the structure. However,
the structure-borne noise or vibration level transmitted to other spaces signi"cantly
depends on the vertical force (normal to the #oor) which generates the bending wave in the
structure. Also, the previous studies concentrated on the system with a single frequency
excitation due to a rotating unbalance. However, the excitation force of an engine is
normally complicated and with multi-frequencies excitation. It can be anticipated that the
optimization results should be quite di!erent.
In the present study, the vibration isolation system of an engine is optimized by using
sequential quadratic programming (SQP) technique. The mounting sti!ness and
orientation of isolators are chosen as the design variables. The objective function to be
minimized is the summation of the transmitted force normal to the base at all mounting
positions. To avoid the selection with excitation frequencies too near to the system
resonance frequencies, a minimum gap between them is set as a constraint condition
in the optimization. The other constraint conditions in the optimization are set to limit
the static and dynamic response of isolators under the excitation and also to meet the
balancing requirement of the engine installation. An example of a 4-stroke engine
with one cylinder is used to show that the optimization technique is e!ective.
The optimization results are compared with that of a conventional engineering design
with all isolators working under their maximum allowable static de#ection. The mechanism
of the vibration attenuation by optimization is investigated based on the frequency
response of objective function and compared with that of the conventional design.
Sensitivity study of the system with the variation of the design parameters is also carried
out and it proves that the minimum obtained is the global value in the range of bounds
of design parameters.
 DESIGN OPTIMIZATION OF MARINE ENGINE-MOUNT SYSTEM 479
2. ENGINE}MOUNT SYSTEM EQUATIONS

The engine is modelled as a rigid body that is supported by four vibration isolators "xed
to a rigid #oor as showed in Figure 1. A rigid-body model is suitable for a structure whose
geometry points remains "xed relative to one another [2]. The right-hand global co-
ordinate system Gxyz has its origin at the centre of mass of the engine when in static
equilibrium. The three orthogonal co-ordinate axes, which are shown in Figure 1, are set
with >, Z-axis parallel to the #oor and X normal to the #oor. The crank-shaft of engine is in
the direction of Z-axis. The rigid-body model consists of six DOFs that include three
translation and three rotation modes respectively. Under the assumption of `smalla motion,
the engine}mount system equation can be written as

[M]+xK ,#[K]+x,"+F,e SR, (1)

where [M] is the 6;6 engine's rigid mass matrix; +x2,"[x y z h h h ] is the
E E E V W X
displacement vector at c.g of engine, [K] is the 6;6 sti!ness matrix, +F, is the 6;1 vector
of excitation forces and moments and u is the forcing angular frequency.
The majority of mounts used in the marine engine mounting are of a rubber bonded to
metal, or elastomeric construction. Complex spring sti!ness is used to model the dynamic
behaviour of the mount [3]. The complex sti!ness of a mount in the three directions of its
local co-ordinate system is de"ned by the equation

[k]"[k](1#jg), (2)

where g is the loss factor and j"(!1. The sti!ness matrix must be transformed from its
local mount co-ordinate system to the global co-ordinate system situated at the engine c.g.
The sti!ness in the global co-ordinate system can be expressed as

+k ,"[A]+k,[A\], (3)
G
where [A] is transpose matrix of the Euler angle matrix [7] and can be written as

 
cos a cos b !sin a cos c#cos a sin b sin c sin a sin c#cos a sin b cos c
[A]" sin a cos b cos a cos c#sin a sin b sin c !cos a sin c#sin a sin b cos c . (4)
!sin b cos b sin c cos b cos c

Figure 1. Rigid body on compliant mounts. The origin of global co-ordinate system is at the centre of gravity of
engine. >- and Z-axis are parallel to the #oor and X-axis is normal to the #oor.
480 J. S. TAO E¹ A¸.

In the above transformation, the "rst is a rotation through an angle a about the Z-axis,
followed by the second rotation through an angle b about >-axis, and then "nal rotation
through c about X-axis.
The sti!ness matrix [K] can be calculated [6] based on the sti!ness matrix [k ] in
G
equation (3) and position matrix [r ] of individual mount i. The displacement vector +x, in
G
equation (1) is thus expressed as

+x,"+F,/([K]!u[M]). (5)

The displacement at each mount is

; "x #X h !> h , (6)

VG E G W G X
; "y !Z h #X h , (7)
WG E G V G X
; "z !X h #> h , (8)
XG E G W G V
where X , > , and Z are the co-ordinates of ith mount at global co-ordinate system.
G G G
The force transmitted to the #oor at each mount in three directions can be written as

+F ,"![k ]+U ,. (9)

G G G
The total force transmitted to the #oor at all mounts is

,
FM " (FM  #FM  #FM  ). (10)
VG WG XG
G
The total force transmitted to the #oor in the vertical direction of all mounts is

,
FM " "FM ". (11)
V VG
G
F and F are the force under the excitation angular frequency u. In the case of excitation
V
force with multi-frequencies (u ), the force F and F are expressed as
H V
L
FM " FM , (12)
SH
H
L
FM " FM . (13)
V VSH
H

3. FORCE AND MOMENT EXERTED BY THE ENGINE

The shaking force and moment of an engine have been detailed investigated by Paul [8].
A single cylinder engine model of Paul is illustrated in Figure 2. After balancing of the
rotating mass and with constant crank angular velocity u, for an engine with one cylinder,
there have only inertia force in the vertical direction and rolling moment in the crank-shaft
 DESIGN OPTIMIZATION OF MARINE ENGINE-MOUNT SYSTEM 481

Figure 2. Single cylinder engine model.

direction. The force and moment in its local co-ordinate system (X, >, Z) as shown in
Figure 2 can be expressed as

F "(m #m )Ru#m Ru(A cos 2h!A cos 4h#A cos 6h!2), (14)
V PMR PCA PCA   
M "(P#m sK )s tan u, (15)
X PCA
where m is the rotation mass, m is the reciprocating mass, s, h, R and u are illustrated in
PMR PCA
Figure 2. P is the gas force inside the cylinder that can be calculated based on the gas
pressure and compression ratio of engine [8]. Typical frequency spectra of F and M are
V X
calculated and illustrated in Figure 3 where (a) is the vertical inertia force and (b) is the
rolling moment. It is found that the disturbing frequencies are at f and 2f for inertia force in
Figure 3(a) but at 1/2f, f, 3/2f and 2f for rolling moment in Figure 3(b). Here, f is the running
frequency of engine which equals to angular velocity u divided by 60.
The forces and moments above are based on the local co-ordinate system on the
crank-shaft and cylinder as shown in Figure 2. These forces and moments should be
transformed to the values in the global co-ordinate system with the origin at the centre of
gravity of engine. The force transformation can be expressed as

 
F F
V V
F "[A\] F , (16)
W W
F F
X X
where [A] is transformation matrix as de"ned in equation (4).
482 J. S. TAO E¹ A¸.

Figure 3. Spectra of excitation force and moment of a 4-stroke engine with one cylinder. (a) is the spectrum of
moment around Z-axis. (b) is the spectrum of force in the X direction.

The moments generated by the force F , F , F in the global co-ordinate systems can be
V W X
expressed as

 
Fy !Fz
X  W 
Mo "ro ;Fo "[r]ro ;[F]ro "([A\][r ])ro ;([A\][F])ro "[A\] F z !Fx ro , (17)
D % %  % % V  X  %
Fx !F y
W  V 
where ro "[io , jo , ko ] is the unit vector of global co-ordinate system. ro "[x, y, z]ro is the
% %
displacement vector from the origin of global co-ordinate system to the point of force.
 DESIGN OPTIMIZATION OF MARINE ENGINE-MOUNT SYSTEM 483
Fo "[F , F , F ]ro is the force vector in the global co-ordinate system. [r ]"[x , y , z ]2
V W X %    
is the translational displacement matrix from the origin of global co-ordinate system to that
of the local co-ordinate system since the force point is considered at the origin of the local
co-ordinate system. The transformation of the moments in the local co-ordinate system to
that in the global co-ordinate system can be expressed as


M
V
o "ro ;Fo "[r]ro ;[F]ro "([A\][r])ro ;([A\][F])ro "[A\] M ro , (18)
M
+ % % % % W %
M
X
where [r] is the displacement matrix from the origin of local co-ordinate system to the
point of force in the local system. In the case that [A] equals a unity matrix where no
rotation exists between two co-ordinate systems, the moments keep the same after the
transformation.
Figure 4(a}f ) illustrates the spectra after the transformation of the force and moment in
Figure 3. The rotation angles a, b, c equal to 53, 103 and 153 respectively. The translation
distances between two co-ordinate systems are 0)02 m, 0)03 m and 0)1 m respectively.

4. OPTIMIZATION

Optimization problem is concerned with the minimization of a objective function F(a)

that may be subject to a number of constraints or bounds. It can be expressed as

Minimize F(a) (19)

subject to

G (a)"0 i"1,2, m, equity constraint, (20)

G
G (a))0 i"m#1,2, n, inequity constraint, (21)
G
a )a)a , bounds of variables, (22)
J S
where F(a) is the objective function, a is the design parameter vector, a is the lower bound of
J
variables while a is the upper bound of the design variables and n is the total number of
S
constraints.
The objective function here is the total force in the direction normal to the installation
base rather than the total force in all directions as applied by Ashra"uon [4] and Snyman
et al. [5]. This is because only the force normal to the base can excite the bending wave
which contributes most of energy of structure-borne noise.
The inequity constraint, called &&dynamic constraint'', is imposed to limit the maximum
allowable dynamic displacement at each mount in three directions and expressed as

"U ")c , (23)

G 
where U is the displacement vector at ith mount and c is the &&dynamic constraint''. This is
G 
critical at low frequency where a large displacement happens. The limits on the
484 J. S. TAO E¹ A¸.

displacement of the c.g of engine as applied by Ashra"uon [4] cannot re#ect the situation at
each mount.
It is possible that the optimization result may be at the point where the system resonance
frequency is just beside the excitation frequency. This often happens at the anti-resonance
point. In this case, a small change of design parameters may cause the ampli"cation of
excitation force rather than the attenuation. Thus, another inequity constraints, called
&&frequency constraint'', must be imposed to set a minimum gap between excitation
frequencies and system resonance frequencies and expressed as

" f !f ")c , (24)

H L 
where f is the selected excitation frequency, f is the natural frequency vector, c is the
H L 
&&frequency constraint'', c should not be too large otherwise it is di$cult to "nd a feasible

solution with optimization, and also not all the excitation frequencies are selected as f .
H
Normally, f is selected at two or three excitation frequencies where with most of excitation
H
energy.
The equity constraints are imposed to limit the orientation of isolator. It is used to keep
as best the same static loading of all mounts in three directions. This will decrease the risk of
losing balance of engine under the strong impact.
The design variables under present study include the sti!ness and orientation of the
isolator. The lower bound a for sti!ness is imposed to limit the maximum allowable static
J
de#ection. This is critical for the stability of engine under the impact force. The upper bound
a for sti!ness is imposed to control the &&wave'' e!ect of isolation system [6]. This is
S
because, as the sti!ness of the mounting system increases, so does the natural frequencies of
the system, it will become more apparent that the system behaves like a distributed system.
The lower bound for the orientation angle of isolator is zero, which means the coincidence
of local co-ordinate system with global co-ordinate system. The upper bound of orientation
angle is 903 that means the isolator is always under compression rather than suspension.
This is the requirement of most of the marine engine installation.
To the optimization algorithms, the genetic algorithms, which were recently introduced
for optimization [9, 10], were considered in the beginning. The big advantage of this
method is that it undertakes a wider search in the entire design variable space than the
conventional gradient-based algorithms. This is mainly due to the random character of the
procreation process in the genetic operators. This wider searching increases the probability
of converging to a global minimum. However, due to the random character of search
process, genetic optimization requires a huge amount of useless objective function
evaluations and thus a long computer time. Fonseca [11] indicated that the best
gradient-based algorithms even produces a better and faster solution than the best genetic
algorithm in a similar amount of computational time. Pogu et al. [12] considered genetic
algorithm and other random search technique only as &&last resort'' methods for solving
continuous optimization problems. Sequential quadratic programming algorithm is
considered as a superior method in the conventional gradient-based algorithms for solving
non-linear constrained optimization problems in terms of computer time and number of
function evaluations [13, 14] and thus is selected in the present study. This technique is also
selected by some distinguished commercial software like Pantran/Nastran [15] and Matlab
[16]. In SQP, the objective and constraint functions are approximated using Taylor series
approximations. However, a quadratic, rather than a linear approximation of objective
function is used which also becomes the part of the name of SQP. The optimization
problem is thus approximated by a quadratic programming (QP) problem at each iteration.
This QP sub-program is solved using a standard QP solver. The problem terminates if the
 DESIGN OPTIMIZATION OF MARINE ENGINE-MOUNT SYSTEM 485
minimum is reached and all constraints are satis"ed. If not, the solution of QP guarantees
that the further descend is possible and approximation process repeats. An overview of SQP
is found in Fletcher [17].
SQP method may not able to "nd a global minimum. This is due to the factors that all the
conventional gradient-based algorithms are based upon the functions with continuous "rst
and second derivatives of objective function and non-linear constraints. However, the
objective function F here may not be able to satisfy this requirement since the absolute
V
value is applied for F in equation (11). Another reason is that the objective function may
V
not be able to satisfy the conditions of strictly convexity and each equity or inequity
constraint may not be a concave function [18]. Thus, the optimization results would
depend on the starting points of variables.
In the current situation, there exist a number of local minima from di!erent starting
points since the space with six-design variables is sliced by constraints into many feasible
and infeasible regions and there is at least one minimum between any two of six natural
frequencies. How to select the starting point to "nd at maximum probability the global
minimum is critical. Tremendous research works have been done in this area due to its
complexity and the achievements have been summarized by Pinter [19] up to 1996. All the
methods have a certain requirement for the objective function, typically a continuously
di!erentiable function. No method is found to be suitable for the current situation with
discontinuously di!erentiable objective and constraint functions. This drawback can only
be overcome by selecting multiple starting points based on the engineering judgement
which also suggested by Fonseca [11]. The study is thus carried out to investigate how to
select the starting points for each variable so as to lead at maximum probability a global
minimum.
At "rst, a program is developed to carry out optimization with simultaneously selected
m equidistant points for each variable between their lower bound a and upper bound a .
H J S
For saving computational time, the termination tolerances for variables, objective and
constraint functions in the optimization are set at higher values, e.g., 10\ for variables and
objective function and 10\ for constraint function, and the evaluation times of objection
function for each starting point is limited within 100. The programme will calculate the
directional derivative in the search direction and compare it with the tolerances of variables
and objective function. The optimization will terminate provided the directional derivative
is quite below these tolerances (at least two times below) and all the constraints are satis"ed.
The optimized results (objective function, variable values at starting points and after
optimization) are saved as long as the optimized objective function value is less than that
with variables at their lower bound a . This implies to search all the solutions better than the
J
conventional engineering design. If the optimized objective function value is found to be the
same as anyone saved, its results will be ignored. In the case that the termination tolerances
have not been satis"ed after 100 times function evaluation, the last optimized results are
also saved provided its function value is less than that at lower bounds of variables.
It is applied in the second step for another programme to do optimization at all saved
starting points. Since the number of saved points is limited, the termination tolerances for
objective and constraint functions in the optimization are reduced down to 10\ and the
evaluation times of objection function for each starting point are enlarged to 300 or higher
to re"ne the "nal minimum from all local minima. This "nal minimum cannot be
guaranteed the global minimum. Theoretically, if m is larger enough, the global minimum
H
can be achieved. But the increase of m will signi"cantly increase computational time.
H
The following strategies can be applied to reduce the number of the starting points, which
comes from the analysis of optimization results: One is to "x the starting point at the lower
bound a for the sti!ness in one direction if the excitation force in this direction is
J
486 J. S. TAO E¹ A¸.

signi"cantly larger than that in other directions ('2 times). This can reduce variables from
six dimensions to "ve dimensions. Another one is to select the second starting point of
sti!ness far from its "rst starting point at lower bound a since the optimized isolator
J
sti!ness always converges to its lower bound as it is near a . This can be understood since
J
the isolation system normally has a good performance at the lowest value of sti!ness. The
last one is to select relatively less starting points for angles since the optimization results are
not sensitive to them. The typical values of angles are selected at 0, 45, 903. It is found that
5}8 starting points for each sti!ness and 3 for each angle can obtain a good result. The
further increase of points cannot improve much of the "nal result. This result should be the
global value in most cases but not guaranteed.

5. NUMERICAL EXAMPLE

Consider a 4-stroke marine engine with one cylinder that is installed as illus-
trated in Figure 1. The engine mass is 9600 kg. The mass moment of inertia
I"[I I I I I I ]"[15 100 15 999 6399 2000 900 1200] kg m. The damping
VV WW XX VW VX WX
loss factor g is 0)1.
The forces and moments in Figure 4 are selected as the input force. The engine is
supported at four corners by four identical isolators. The location of each mount is listed in
the Table 1. Obviously, the c.g. of engine is not at the centre of the geometry of engine block.
The lower and upper bounds of isolator sti!ness are

1)92e#6 N/m)(K , K , K ))3)0e#8 N/m. (25)

V W X
Based on this lower and upper bounds, the maximum allowable static de#ection for an
isolator in three directions is 0)0125 m (0)5 in) which is a typical value of neoprene isolator,
and the maximum system frequency is about 125 Hz. Bolton-Knight [20] indicated that
when an isolator excited at frequencies above 200 Hz, standing waves within the rubber
would become signi"cant.
The &&dynamic constraint'' is set to 0)005 m (0)2 in) which is a typical value of neoprene
isolator. With this constraint, the isolator will work within the allowable range of its
de#ection. In the case the isolator is overloaded in the static equilibrium (the de#ection of
isolator is over its rated static de#ection), the dynamic constraint should be more stringent.
The &&frequency constraint'' is set to 5 Hz. The main excitation frequencies f selected in
H
the &&frequency constraint'' are 25 and 50 Hz which with most of excitation energy.
The orientation of each isolator is set such that, in each direction, the absolute value of
angle of each mount is the same but it may be arranged in the negative direction to satisfy
the balance requirement of engine installation. The arrangement of four mount orientations
should meet the following equity constraints:

!a "a "a "!a , (26)

   
b "b "!b "!b , (27)
   
c "c "c "c , (28)
   
where a means the angle around positive Z-axis at location 1 as shown in Figure 1.

There are totally 24 inequity constraints and 9 equity constraints in this example. The
original design is to install isolators under their maximum allowable static de#ection
 DESIGN OPTIMIZATION OF MARINE ENGINE-MOUNT SYSTEM 487

Figure 4. The spectra of the forces and moments after the transformation of the force and moment in Figure 3.
The translational displacement of origin of local co-ordinate system is (0)02 m, 0)03 m, 0)1 m) and the rotation
angles are (5,10,15) deg.

(0)0125 m) and the isolator's principle elastic directions coincide with the global co-ordinate
system (a, b, c"0). This is a typical engineering design with the full usage of the potential of
isolators.

6. RESULTS AND DISCUSSION

After being optimized, the vertical force transmitted to the base can be reduced to about
only half of its original design. Table 2 gives the comparison between the original and
488 J. S. TAO E¹ A¸.

TABLE 1
Mounting locations of isolators

X(m) >(m) Z(m)

c.g. 0 0 0
Mount 1 !1 0)8 2)2
Mount 2 !1 !1)2 2)2
Mount 3 !1 !1)2 !1)8
Mount 4 !1 0)8 !1)8

TABLE 2
Comparison of the objective function and design variables between original
and optimized systems

Original system Optimized system

Objective function (F , N) 212 117

V
K (N/m) 1 920 000 1 920 000
V
K (N/m) 1 920 000 9 460 000
W
K (N/m) 1 920 000 45 300 000
X
a(Z direction, deg) 0 4
b(> direction, deg) 0 11
c(X direction, deg) 0 90

TABLE 3
Comparison of the system natural frequencies between original and optimized systems
Original system natural Optimized system Excitation
frequencies (Hz) natural frequencies (Hz) frequencies (Hz)

2)9 4)2 12)5

3)8 5 25
4)5 5)3 37)5
7)7 13)4 50
7)8 32
9)6 40)4

optimized systems for objective function, isolator sti!ness and mounting orientations.
Table 3 gives the comparison of system natural frequencies between the original and
optimized systems as well as excitation frequencies. The bolded frequencies are that
with main energy of excitation and selected in the &&frequency constraint'' for optimization.
Table 4 gives the comparison between the original and optimized systems for the
displacement of engine c.g. It is found from Table 2 that the isolators are selected with
higher sti!ness in the optimized system. This implies that the high installation de#ection in
the original system do not mean the higher isolation performance. The careful selection and
installation of isolators may get better results even with the lower static de#ection. On the
 DESIGN OPTIMIZATION OF MARINE ENGINE-MOUNT SYSTEM 489
TABLE 4
Comparison of de-ection of c.g. of engine between original and optimized systems

De#ection Original system Optimized system

x (m) 2)24;10e!5 2)16;10e!5

E
y (m) 2)1;10e!7 2)26;10e!6
E
z (m) 6;10e!6 8;10e!6
E
h (deg) 8)6;10e!5 2)3;10e!4
V
h (deg) 1)5;10e!4 1)3;10e!4
W
h (deg) 3)9;10e!4 1)1;10e!4
X

other hand, the reduced static de#ection of isolator would also be bene"cial to the dynamic
balancing of engine mounting system.
The mechanism of optimization was investigated. It can be seen from Table 3 that the
isolation improvement in the optimized system is not by moving a natural frequency from
an input forcing frequency so as to increase the frequency ratio f/f as stated by Swanson [3]

and Spiekermann [2]. This is because the original isolation system has lower system natural
frequencies and the higher frequency ratio. From Table 4, it seems that this improvement is
achieved by obtaining the larger displacement of c.g. of engine in the > direction since the
value in this direction increases about 10 times. The further investigation is carried out by
analyzing the system frequency response with the optimized variables in Table 2. The forces
and moments at excitation frequency 25 Hz is selected from all the excitation frequencies as
the input force matrix +F, in equation (1) since most of excitation energy is at this frequency.
Figure 5 shows the frequency response of original and optimized systems under the same
excitation force. It can be seen from Figure 5(a) that the mechanism of vibration isolation of
the original system is by moving system natural frequencies quite below the excitation
frequencies. This is the conventional design guide for the isolation system. It requires the
high static de#ection and large allowable dynamic range of movement of isolators and
engine block. However, the optimization method is by "nding the force dips at excitation
frequencies as showed in Figure 5(b). Compared with the force values at 25 Hz in
Figure 5(a, b), it is found that the value of optimized system at this frequency is only about
1/4 of the original system. However, this value is about 1/2 (refer to Table 1) for multi-
frequencies excitation. It proves that optimization is more e!ective for a system with a single
excitation frequency.
In any constrained optimization, it is important to see which constraints limit the
performance of the system. These constraints are called active constraint. Relaxation of
these constraints would increase the system isolation performance. On the other hand, the
relaxation of other constraints may have no e!ect on the optimal solution. Table 5 presents
the dynamic de#ection in three directions at each mount. Since all the values in Table 5 are
much lower than the &&dynamic constraint'' of 5 mm, it is obvious that relaxation of this
constraint would have no e!ect on the system performance. However, the relaxation of
&&frequency constraint'' 5 Hz would have signi"cant impact to the system performance. This
impact is shown in Table 6 that the isolation system is optimized with &&frequency
constraint'' at 0, 5 and 10 Hz. The optimized design variables at each case are also given in
the table. It is clearly indicated that, without &&frequency constraint'', the system has the best
isolation performance. As &&frequency constraint'' increases from 5 to 10 Hz, the transmitted
force of the optimized systems increases from 117 to 146 N. However, this value only
490 J. S. TAO E¹ A¸.

Figure 5. Illustration of the spectra of frequency response of the original (a) and optimized (b) systems. The input
force matrix +F, is at 25 Hz with most of energy of excitation.

increases about 3 N as &&frequency constraint'' varies from 0 to 5 Hz. This implies that
&&frequency constraint'' can be carefully selected to reduce the possibility of system
resonance but keep almost the same system isolation performance. As the &&frequency
constraint'' is further increased from 10 Hz, it can be anticipated that the transmitted force
also becomes larger and gets close to the value of conventional design (212 N). Another
phenomenon found from Table 6 is that, in three cases, K , K and c are almost the same
V W
after changing &&frequency constraint'', a and b with small adjustment, but K is signi"cantly
X
 DESIGN OPTIMIZATION OF MARINE ENGINE-MOUNT SYSTEM 491
TABLE 5
Dynamic de-ection in three directions at each mount

Point 1 Point 2 Point 3 Point 4

; (mm) 0)02602 0)02965 0)02063 0)01873

V
; (mm) 0)01167 0)01167 0)00800 0)00800
W
; (mm) 0)00800 0)01540 0)01540 0)00800
X

TABLE 6
Comparison of the objective function and design variables between optimized systems with
di+erent 00frequency constraint11

Optimized system Optimized system Optimized system

with &&frequency with &&frequency with &&frequency
constraint'' 0 Hz constraint'' 5 Hz constraint'' 10 Hz

Objective function (F , N) 114 117 145

V
K (N/m) 1 920 000 1 920 000 1 920 000
V
K (N/m) 9 700 000 9 460 000 9 490 000
W
K (N/m) 33 700 000 45 300 000 192 600 000
X
a 3)6 4 11)6
b 8)8 11 13)8
c 90 90 90

Figure 6. Illustration of the spectra of frequency response of the systems without &&frequency constraint'' (solid
line), 5 Hz &&frequency constraint'' (dotted line) and 10 Hz &&frequency constraint'' (dash}dotted line). The input
force matrix +F, is at 25 Hz with most of the energy of excitation.
492 J. S. TAO E¹ A¸.

changed. It seems that K works as a tuner of the system natural frequency so as to meet the
X
&&frequency constraint''. Figure 6 illustrates the frequency response of system under three
di!erent &&frequency constraints''. It shows that as &&frequency constraint'' reduces from 10 to
0 Hz, one of natural frequencies of optimized system will get closer and closer to one of the
excitation frequencies (25 Hz). This implies the higher and higher possibility of system
resonance. It can thus be concluded that &&frequency constraint'' should be selected with
a compromise between high isolation performance and possible system resonance. It is also
interesting to "nd that, at &&frequency constraint'' 10 Hz, the transmitted force of optimized
system has the lowest value at 25 Hz in Figure 6 but the function value at this &&frequency
constraint'' is the highest in Table 6 as the system has higher force transmission in other
excitation frequencies. This implies that the objective to minimize the force transmission at
the engine operating frequency (25 Hz at 1500 r.p.m.) does not mean the best design of the
isolation system.

Figure 7. Illustration of system response of one design parameter varying from its lower bound to its upper
bound. (a) is with K varying, (b) is K , (c) is K , (d) is a, (e) is b, (f) is c.
V W X
 DESIGN OPTIMIZATION OF MARINE ENGINE-MOUNT SYSTEM 493
Since SQP method may not be able to "nd a global minimum, it is necessary to
investigate whether there is any other solution that has the lower value of objective function.
This has been done by adjusting a design variable from its lower bound to its upper bound
while keeping other variables at optimal values. Figure 7(a}f ) illustrates the system response
at &&frequency constraint'' 5 Hz to the variation of all design parameters including
K , K , K , a, b, c. It can be seen that the lowest points of objective function are at the
V W X
optimal values of variables. Figure 7 also illustrates the system sensitivity to the variation of
the design parameters. It can be seen from Figure 7 that the objective function is not quite
sensitive to the angles but it is sensitive to sti!ness. Due to the large scale of horizontal axis
(3;10), the objective function should not increase too much for a certain percent variation
of the sti!ness. The calculated force variation is 13 N for 10% change of K , 16 N for K and
V W
24 N for K . It can be anticipated that the relaxation of &&frequency constraint'' can reduce
X
the force transmission but the force variations above will increase.

7. CONCLUSION

The numerical optimization of a typical 4-stroke engine with one cylinder is presented.
The dynamic force in the direction normal to the installation base is minimized. The design
parameters are the sti!ness coe$cients and orientation angles of mounts. Two constraints
are imposed in the system. One is to keep the engine static and dynamic de#ection within
the desired limits. The other one is to set a minimum gap between system natural frequency
and engine excitation frequency. The SQP technique has been employed successfully for the
optimization. The results of optimization are compared with that of a typical engineering
design with the isolators working under their maximum allowable defection. It is found that
the transmitted force of the optimized system is signi"cantly reduced. The mechanism of
vibration isolation involved in the optimization is investigated with the frequency response
of system. It shows that, di!ering from the traditional isolation system design by shifting the
system natural frequencies to lower value, the optimized system is to reduce the force
transmission by searching the force dip in the system response to all design variables. The
optimization results also indicate that the force transmission for a system with a single
excitation frequency is much lower than that for a multi-frequencies excitation. Sensitivity
study of the system with the variation of the design parameters is also carried out and it
proves that the minimum obtained is the global value in the range of design parameters.

REFERENCES

1. E. SEVIN and D. P. WALTER 1971 Optimal Shock and <ibration Isolation. The Shock and
Vibration Information Centre, United States Department of Defence.
2. C. E. SPIEKERMANN, C. J. RADCLIFFE and E. D. GOODMAN 1985 Journal of Mechanisms,
¹ransmissions, and Automation in Design 107, 271}276. Optimal design and simulation of
vibration isolation system.
3. D. A. SWANSON, H. T. WU and H. ASHRAFIUON 1993 Journal of Aricraft 30, 979}984.
Optimisation of aircraft engine suspension systems.
4. H. ASHRAFLIUON 1993 Journal of <ibration and Acoustics 115, 463}467. Design optimisation of
aircraft engine-mount systems.
5. J. A. SNYMAN, P. S. HEUNS and P. J. VERMEULEN 1995 Mechanical and Machine ¹heory 30,
109}118. Vibration isolation of a mounted engine through optimisation.
6. C. M. HARRIS and C. E. CREDE 1976 Shock and <ibration Handbook. New York: McGraw-Hill,
second edition.
7. D. T. GREENWOON 1979 Principles of Dynamics. Englewood Cli!s, NJ: Prentice-Hall, second
edition.
494 J. S. TAO E¹ A¸.

8. B. PAUL 1979 Kinematics and Dynamics of Planar Machinery. Englewood Cli!s, NJ: Prentice-Hall.
9. D. E. GOLDBERG 1989 Genetic Algorithms in Search, Optimisation and Machine ¸earning. New
York: Addison-Wesley.
10. K. S. TANG, K. F. MAN, S. KWONG and Q. HE 1996 Proceedings of the Fourth International
Congress on Sound and <ibration. I, 371}388. Application of genetic algorithms to active noise and
vibration control.
11. P. DE FONSECA, P. SAS and H. VAN BRUSSEL 1999 Journal of Sound and <ibration 221, 651}679.
A comparative study of methods for optimising sensor and actuator locations in active control
application.
12. M. POGU and J. E. SOUZA DE CURSI 1994 Journal of Global Optimisation 5, 159}180. Global
optimisation by random perturbation of the gradient method with a "xed parameter.
13. G. N. VANDERPLAATS and H. SUGIMOTO 1985 Engineering Computations 2, 96}100. Application
of variable metric methods to structural synthesis.
14. M. J. D. POWELL 1978 Mathematical Programming 14, 224}248. Algorithms for non-linear
constraints that use Lagrangian functions.
15. G. J. MOORE 1994 Design Sensitivity and Optimisation, MSC/Nastran ;ser's guide, Vol. 68. The
Macneal-Schwendler Corp.
16. T. GOLEMAN, M. A. BRANCH and A. GRACE 1999 Matlab Optimisation ¹oolbox, ;ser's Guide,
<ersion 2. The Mathworks Inc.
17. R. FLETCHER 1980 Practical Methods of Optimisation. New York: John Wiley and Sons.
18. D. M. HIMMELBLAU 1972 Applied Nonlinear Programming. New York: McGraw-Hill.
19. J. D. PINTER 1996 Global Optimisation in Action. Dordrecht: Kluwer Academic Publishers.
20. B. L. BOLTON-KNIGHT 1971 Institution for Mechanical Engineers C98, 24}31. Engine mounts:
analytical methods to reduce noise and vibration.